portal-vlm: qwen25vl-hypernet

PorTAL hypernet (task latent + decoder core + Qwen2.5-VL converter) generating r8 LoRA for 252 LM sites. The SOURCE checkpoint both ported rows were refit from.

Part of the portal-vlm release - an independent replication of Ramp Labs' PorTAL (portable task adapters via hypernet-generated LoRA) extended to vision-language models on GUI grounding.

	ScreenSpot-v2 overall	web split
this artifact	82.8%	79.9%

Base model: Qwen/Qwen2.5-VL-3B-Instruct @ 66285546d2b821cf421d4f5eb2576359d3770cd3
Training config: configs/qwen25vl_hypernet.yaml
Eval record: results/qwen25vl_hypernet.json

Reproduce this row without training

git clone https://github.com/robbym-dev/portal-vlm && cd portal-vlm && uv sync
uv run python scripts/eval.py --config configs/qwen25vl_hypernet.yaml --adapter hf:manihani4/portal-vlm-qwen25vl-hypernet

hypernet.pt is a PorTAL hypernet checkpoint (task latent + decoder core + per-base converters), not a PEFT adapter: at eval time the hypernet generates r8 LoRA factors for the target base, which scripts/eval.py injects on the fly.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for manihani4/portal-vlm-qwen25vl-hypernet

Base model

Qwen/Qwen2.5-VL-3B-Instruct

Adapter

(276)

this model